bitsnbytes - AshokBhat/ml GitHub Wiki

About

  • Lightweight Wrapper: Simplifies using CUDA custom functions in Python.
  • 8-Bit Optimizers: Supports optimizers like Adam and RMSProp.
  • Quantization Support: Facilitates 4-bit and 8-bit operations.
  • Multi-Backend Compatibility: Works with Intel, AMD, and Apple Silicon.
  • Memory Efficiency: Reduces memory usage significantly during training

See also